Summary of Continual Learning Of Large Language Models: a Comprehensive Survey, by Haizhou Shi and Zihao Xu and Hengyi Wang and Weiyi Qin and Wenyuan Wang and Yibin Wang and Zifeng Wang and Sayna Ebrahimi and Hao Wang
Continual Learning of Large Language Models: A Comprehensive Survey
by Haizhou Shi, Zihao Xu, Hengyi Wang, Weiyi Qin, Wenyuan Wang, Yibin Wang, Zifeng Wang, Sayna Ebrahimi, Hao Wang
First submitted to arxiv on: 25 Apr 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The recent success of large language models (LLMs) has sparked numerous research directions and applications. This survey provides a comprehensive overview of the current research progress on LLMs within the context of continual learning, which addresses the challenge of integrating pre-trained LLMs into dynamic data distributions, task structures, and user preferences. Continual Pre-Training (CPT), Domain-Adaptive Pre-training (DAP), and Continual Fine-Tuning (CFT) are three stages of learning LLMs in the context of modern CL. The survey also discusses evaluation protocols for continual learning with LLMs, along with the current available data sources. This survey is structured into four main sections: an overview of continually learning LLMs, comprising two directions of continuity; a summary of the three stages of learning LLMs in the context of modern CL; an overview of evaluation protocols and data sources; and a discussion of intriguing questions pertaining to continual learning for LLMs. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Large language models (LLMs) have been very successful when trained on general datasets. But what happens if we want to use these models on new, specific tasks or domains? This is called “catastrophic forgetting.” The research community has been working on a solution called continual learning (CL), where the model adapts to new tasks and data without losing its old skills. In this survey, we look at how LLMs can be used in CL. We see that there are different ways to do this, such as Continual Pre-Training (CPT), Domain-Adaptive Pre-training (DAP), and Continual Fine-Tuning (CFT). The survey also talks about how we evaluate these models and what kind of data is available. The main question the survey answers is: How can we use LLMs in CL to make them learn new things without forgetting old skills? |
Keywords
» Artificial intelligence » Continual learning » Fine tuning
